Recently, soft sensors have been widely adopted as alternatives to physical sensors. However, their generalization performance often degrades under distribution shifts, and their prediction errors are not directly observable without ground-truth measurements. As a result, cost-expensive sensors that consume significant energy or resources are still required in practice. To address this challenge, we study cost-efficient sensing scheduling in systems equipped with both soft and cost-expensive sensors, operating under a limited budget for activating the latter. We first develop an optimal sensing scheduling algorithm that minimizes measurement error while satisfying an average cost constraint. To enable its practical deployment, we propose a self-improving soft sensor that estimates predictive uncertainty and enhances its performance using selectively collected ground-truth measurements. We then design a cost-efficient sensing scheduling framework by integrating them. In this framework, predictive uncertainty serves as a proxy for unknown measurement errors, which provides guidance on when to activate cost-expensive sensors. Based on this guidance, the scheduling algorithm selectively acquires informative measurements that facilitate not only error reduction but also soft sensor improvement, given the limited budget. Extensive simulations on real-world datasets demonstrate that the proposed framework effectively quantifies predictive uncertainty, improves soft sensor accuracy, and significantly outperforms baseline algorithms in both error reduction and data efficiency.

In this article, we address an edge computing resource allocation and deep neural network (DNN) splitting problem in an edge–cloud collaborative system to minimize the task execution time of a multisensor digital twin (DT), where the constituent tasks of the multisensor DT are employed by DNN models with both split computing and early exit structures. To this end, we develop an early exiting-aware joint edge computing resource allocation and DNN splitting (ERDS) framework that optimally solves the problem. In the framework, the problem is reformulated into a nested optimization problem consisting of an outer edge computing resource allocation problem and an inner DNN splitting problem which considers early exiting. Based on the nested structure, the framework can efficiently solve the problem without having to consider the ERDS jointly. As components of the framework, we develop an edge computing resource allocation algorithm that exploits the mathematical structure of the outer problem; we also develop an optimal DNN splitting algorithm and a heuristic algorithm that identifies suboptimal solutions but has lower computational complexity. Through the simulation, we demonstrate that our proposed framework effectively outperforms the other state-of-the-art baselines in terms of the task execution time of the multisensor DT in different environments, which shows that our proposed framework is applicable in practical multisensor DTs.

In this article, we examine cloud–edge–terminal Internet of Things (IoT) networks, where edges undertake a range of typical dynamic scheduling tasks. In these IoT networks, a central policy for each task can be constructed at a cloud server. The central policy can be then used by the edges conducting the task, thereby mitigating the need for them to learn their own policy from scratch. Furthermore, this central policy can be collaboratively learned at the cloud server by aggregating local experiences from the edges, thanks to the hierarchical architecture of the IoT networks. To this end, we propose a novel collaborative policy learning framework for dynamic scheduling tasks using federated reinforcement learning. For effective learning, our framework adaptively selects the tasks for collaborative learning in each round, taking into account the need for fairness among tasks. In addition, as a key enabler of the framework, we propose an edge-agnostic policy structure that enables the aggregation of local policies from different edges. We then provide the convergence analysis of the framework. Through simulations, we demonstrate that our proposed framework significantly outperforms the approaches without collaborative policy learning. Notably, it accelerates the learning speed of the policies and allows newly arrived edges to adapt to their tasks more easily.

In this paper, we propose a novel meta-scheduling framework with cooperative learning that fully exploits a functional split structure of the base station (BS) consisting of a central unit (CU) and digital units (DUs). To this end, we first design a meta-scheduling policy structure to find a scheduling strategy that can be applied to any BS regardless of BS-specific characteristics such as the number of users, various quality-of-service requirements, and the number of resource blocks. In the proposed framework, the CU only needs to manage objective-specific meta-scheduling policies in a centralized manner while each DU performs scheduling in a decentralized manner by simply using the policy at the CU without any responsibility to manage or train policies. Besides, the policies at the CU can be cooperatively trained using the experiences obtained from all DUs. Therefore, the proposed framework not only provides computational efficiency but also supports network scalability. We show via experiments that the proposed meta-scheduling framework achieves competitive performances compared with the near-optimal conventional schedulers tailored to each BS even though it manages and exploits only one meta-scheduling policy for each objective. Furthermore, our meta-scheduling framework effectively adapts to the change of BS-specific characteristics.

Dynamic scheduling is an important problem in applications from queuing to wireless networks. It addresses how to choose an item among multiple scheduling items in each timestep to achieve a long-term goal. Most of the conventional approaches for dynamic scheduling find the optimal policy for a given specific system so that the policy from these approaches is usable only for the corresponding system characteristics. Hence, it is hard to use such approaches for a practical system in which system characteristics dynamically change. This paper proposes a novel policy structure for MDP-based dynamic scheduling, a descriptive policy, which has a system-agnostic capability to adapt to unseen system characteristics for an identical task (dynamic scheduling). To this end, the descriptive policy learns a system-agnostic scheduling principle–in a nutshell, “which condition of items should have a higher priority in scheduling”. The scheduling principle can be applied to any system so that the descriptive policy learned in one system can be used for another system. Experiments with simple explanatory and realistic application scenarios demonstrate that it enables system-agnostic meta-learning with very little performance degradation.

In this article, we study the waveform scheduling problem for a wireless power transfer (WPT) system consisting of a power beacon (PB) and multiple energy-harvesting-empowered Internet of Things (EH-IoT) devices. In each time slot, each device requests power to the PB if it needs power, and the PB transmits a WPT signal for which the waveform is designed based on the harvested power satisfaction rate of the power-requesting devices. Under this setup, we formulate an optimization problem that maximizes the average number of EH-IoT devices whose power requests are satisfied. We first solve this problem, assuming that the perfect channel state information (CSI) of all devices is known at the PB. Since the problem is difficult to solve even with perfect CSI, we transform it into a more tractable problem via proper approximations and propose an efficient algorithm to solve it. Next, to tackle the issue that it is practically difficult for the PB to acquire the perfect CSI of each device, we propose a contextual learning-based WPT waveform scheduling algorithm, requiring only 1-bit feedback from each device at one time. Numerical results show that our proposed waveform scheduling algorithm provides a higher satisfaction rate than existing algorithms under perfect CSI, and that with limited CSI feedback achieves performance close to the case with perfect CSI.

In this paper, we study radio and energy resource management in renewable energy-powered wireless networks, where base stations (BSs) are powered by both on-grid and renewable energy sources and can share their harvested energy with each other. To efficiently manage those resources, we propose a hierarchical and distributed resource management framework based on deep reinforcement learning. The proposed framework minimizes the on-grid energy consumption while satisfying the data rate requirement of each user. It is composed of three different policies in a distributed and hierarchical way. An intercell interference coordination policy constrains the transmission power at each BS to coordinate the intercell interference among the BSs. Under the power constraints, a distributed radio resource allocation policy of each BS determines its own user scheduling and power control. Lastly, an energy sharing policy manages the energy resources of the BSs by sharing the harvested energy via power lines between them. Through the simulation, we demonstrate that the proposed framework can effectively reduce the on-grid energy consumption while satisfying the data rate requirements.

Phase I clinical trials are designed to test the safety (non-toxicity) of drugs and find the maximum tolerated dose (MTD). This task becomes significantly more challenging when multiple-drug dose-combinations (DC) are involved, due to the inherent conflict between the exponentially increasing DC candidates and the limited patient budget. This paper proposes a novel Bayesian design, SDF-Bayes, for finding the MTD for drug combinations in the presence of safety constraints. Rather than the conventional principle of escalating or de-escalating the current dose of one drug (perhaps alternating between drugs), SDF-Bayes proceeds by cautious optimism: it chooses the next DC that, on the basis of current information, is most likely to be the MTD (optimism), subject to the constraint that it only chooses DCs that have a high probability of being safe (caution). We also propose an extension, SDF-Bayes-AR, that accounts for patient heterogeneity and enables heterogeneous patient recruitment. Extensive experiments based on both synthetic and real-world datasets demonstrate the advantages of SDF-Bayes over state of the art DC trial designs in terms of accuracy and safety.

In this article, we study a random access (RA) scheme to alleviate the RA channel (RACH) overload problem in the massive-machine-type communication (mMTC) environment. We first propose a timing advance-based preamble resource expansion (TAPRE) scheme which effectively increases preamble resources and reduces the preamble collision probability by adjusting preamble transmission timing with timing advance (TA) information. We also propose a resource allocation wait (RAW) scheme which efficiently reduces the number of RA failures due to the lack of physical uplink shared channel (PUSCH) resources. We then propose an overall procedure for enhanced RA with TAPRE and RAW (ERATAR). In addition, we provide the analysis of RA performance by applying more practical assumptions than the existing analysis. We validate our analysis with the system level simulation based on NS-3, and compare the various performances of our ERATAR to those of existing works. Numerical results show that our analysis provides more accurate results than the existing work and our ERATAR provides significantly improved performances compared with those of existing works.

In this paper, we study asynchronous federated learning (FL) in a wireless distributed learning network (WDLN). To allow each edge device to use its local data more efficiently via asynchronous FL, transmission scheduling in the WDLN for asynchronous FL should be carefully determined considering system uncertainties, such as time-varying channel and stochastic data arrivals, and the scarce radio resources in the WDLN. To address this, we propose a metric, called an effectivity score, which represents the amount of learning from asynchronous FL. We then formulate an Asynchronous Learning-aware transmission Scheduling (ALS) problem to maximize the effectivity score and develop three ALS algorithms, called ALSA-PI, BALSA, and BALSA-PO, to solve it. If the statistical information about the uncertainties is known, the problem can be optimally and efficiently solved by ALSA-PI. Even if not, it can be still optimally solved by BALSA that learns the uncertainties based on a Bayesian approach using the state information reported from devices. BALSA-PO suboptimally solves the problem, but it addresses a more restrictive WDLN in practice, where the AP can observe a limited state information compared with the information used in BALSA. We show via simulations that the models trained by our ALS algorithms achieve performances close to that by an ideal benchmark and outperform those by other state-of-the-art baseline scheduling algorithms in terms of model accuracy, training loss, learning speed, and robustness of learning. These results demonstrate that the adaptive scheduling strategy in our ALS algorithms is effective to asynchronous FL.

Clinical trials in the medical domain are constrained by budgets. The number of patients that can be recruited is therefore limited. When a patient population is heterogeneous, this creates difficulties in learning subgroup specific responses to a particular drug and especially for a variety of dosages. In addition, patient recruitment can be difficult by the fact that clinical trials do not aim to provide a benefit to any given patient in the trial. In this paper, we propose C3T-Budget, a contextual constrained clinical trial algorithm for dose-finding under both budget and safety constraints. The algorithm aims to maximize drug efficacy within the clinical trial while also learning about the drug being tested. C3T-Budget recruits patients with consideration of the remaining budget, the remaining time, and the characteristics of each group, such as the population distribution, estimated expected efficacy, and estimation credibility. In addition, the algorithm aims to avoid unsafe dosages. These characteristics are further illustrated in a simulated clinical trial study, which corroborates the theoretical analysis and demonstrates an efficient budget usage as well as a balanced learning-treatment trade-off.

Subgroup analysis of treatment effects plays an important role in applications from medicine to public policy to recommender systems. It allows physicians (for example) to identify groups of patients for whom a given drug or treatment is likely to be effective and groups of patients for which it is not. Most of the current methods of subgroup analysis begin with a particular algorithm for estimating individualized treatment effects (ITE) and identify subgroups by maximizing the difference across subgroups of the average treatment effect in each subgroup. These approaches have several weaknesses: they rely on a particular algorithm for estimating ITE, they ignore (in)homogeneity within identified subgroups, and they do not produce good confidence estimates. This paper develops a new method for subgroup analysis, R2P, that addresses all these weaknesses. R2P uses an arbitrary, exogenously prescribed algorithm for estimating ITE and quantifies the uncertainty of the ITE estimation, using a construction that is more robust than other methods. Experiments using synthetic and semi-synthetic datasets (based on real data) demonstrate that R2P constructs partitions that are simultaneously more homogeneous within groups and more heterogeneous across groups than the partitions produced by other methods. Moreover, because R2P can employ any ITE estimator, it also produces much narrower confidence intervals with a prescribed coverage guarantee than other methods.

In this paper, we study a waveform scheduling problem for a multi-receiver wireless power transfer (WPT) system considering time-varying channel conditions and minimum average output direct-current (DC) voltage requirement of each receiver. To this end, we formulate a stochastic optimization problem that aims at maximizing the average of the sum of output DC voltages of receivers while satisfying the minimum average output DC voltage requirements of all receivers, and by solving it, we develop a waveform scheduling algorithm. In the waveform scheduling algorithm, we need to solve a problem for maximizing the weighted-sum of output DC voltages of receivers, which is a non-convex optimization problem. To cope with this difficulty, we develop a low-complexity approximated algorithm with which the waveform for the multi-receiver WPT system is optimized to maximize the weighted sum of output DC voltages of receivers. Numerical results show that our waveform design algorithm provides the higher performance of the weighted-sum of the output DC voltages than the existing algorithms, and our opportunistic waveform scheduling provides good performance while well satisfying the minimum average output DC voltage requirement of each receiver.

In this paper, we consider Internet of Things (IoT) systems that can be applied to various applications with low-mobility or static IoT devices, such as wireless sensor networks and charging systems for low-power devices with communication. The IoT systems consist of IoT devices and a hybrid access point (H-AP) powered by both on-grid and renewable energy sources. The IoT devices have a capability to harvest energy from the H-AP's radio frequency signal, and they perform their tasks by using only their harvested energy. We consider the tasks do not have a real-time requirement which can be stored in the task queues of the IoT devices and performed later. We study resource and task scheduling for the IoT systems which aims at minimizing the on-grid energy consumption at the H-AP while guaranteeing the minimum average data rate and minimum task performing rates of IoT devices. To achieve the goal, we first propose a centralized resource and task scheduling algorithm. However, its computational complexity and signaling overhead are too large due to the task scheduling for each IoT device. Thus, to resolve these issues, we propose a hybrid resource and task scheduling algorithm in which each IoT device determines its own task scheduling in a distributed manner and the H-AP determines the resource scheduling. We then provide performance analyses showing that our proposed algorithms are asymptotically optimal and well satisfy the QoS requirements of IoT devices even with distributed task scheduling. Through the simulation results, we verify the analyses and show the performance of our algorithms.

In this paper, we consider Internet of Things (IoT) systems in which IoT devices request power to a power beacon (PB) when their available power is deficient and the PB provides power to the IoT devices using switched beamforming. We study wireless power transfer (WPT) beam scheduling for the IoT systems under one-bit feedback which aims at maximizing the time-average number of the IoT devices whose power requests are satisfied. To achieve this, we propose a contextual learning-based WPT beam scheduling algorithm with one-bit feedback (CWBO) that learns the channel information using only one-bit feedback information and exploits it for the beam scheduling. Within CWBO, a beam pattern generation (BPG) problem should be solved in each time slot. To efficiently solve it, we develop a BPG algorithm based on monotonic optimization that can optimally solve the BPG problem. In addition, we also develop a heuristic BPG algorithm that has a lower computational complexity than the monotonic optimization-based BPG algorithm, while providing comparable performance. For CWBO in single-device WPT, we prove an analytical performance bound, which shows its optimality in terms of the long-term average performance even with one-bit feedback. In addition, through the simulation results, we show that our algorithms achieve performances close to that of the optimal beam scheduling policy in multidevice WPT as well. This demonstrates that our algorithms can be used for WPT IoT systems with IoT devices having only limited capabilities for feedback and estimation of the channel information due to their limited power.

In this paper, we study the waveform design in the wireless power transfer (WPT) system with multiple receivers. In the multi-receiver WPT system, due to the severe power attenuation of RF signals according to the distance and the heterogeneity in the energy harvesting capability, there exists severe unfairness in energy harvesting among receivers at different distances from the transmitter and/or different energy harvesting capabilities. Hence, alleviating unfairness in energy harvesting among receivers is one of the critical challenges in the multi-receiver WPT system. To tackle this challenge in a systematic way, we consider the fairness in our waveform design applying the α-proportional fairness. With the analysis of the rectenna circuit, we derive the output dc voltage and power of the rectifier in closed forms. Thereby, we formulate an optimization problem to design the waveform for fair WPT. The problem is shown to be a non-convex optimization problem, which is hard to solve in general. However, with proper approximations, we convert it into a convex optimization problem that can be solved easily and obtain the waveform for fair WPT. Numerical results show that our designed waveform makes receivers harvest energy fairly and can control the degree of the fairness easily.

In this paper, we study a random access (RA) procedure to support the massive connectivity of the Internet of Things (IoT) devices, also known as the IoT connectivity. Compared with the previous RA procedures that have limitations to support the IoT connectivity due to the exponentially increased access delay, we develop an RA procedure by applying the sparse code multiple access to reduce the access delay and increase the ratio of the IoT devices that successfully complete their RA procedures. We provide the theoretical performance analysis of the proposed RA procedure with the performance metrics, such as the RA success probability, the average access delay, the RA throughput, and the average number of preamble transmissions. Then, we provide the numerical results to evaluate the performance of the proposed RA procedure based on our analysis and the ns-3 simulator. Numerical results show that our proposed RA procedure is able to support the massive connectivity requirement with improved RA performance metrics compared with the conventional RA procedures.

In this paper, we study joint energy cooperation and traffic management in renewable energy powered cellular system where a centralized unit manages the traffic and the energy cooperation among BSs. We first formulate a stochastic optimization problem which aims at minimizing the total on-grid energy consumption while satisfying the quality-of-service (QoS) requirement of classes of services, i.e., the minimum average data rates. By using the Lyapunov optimization framework, we develop a joint adaptive energy cooperation and traffic management algorithm which does not need the statistical information of the system. Then, we provide the performance analysis which shows our proposed algorithm is asymptotically optimal. Through the simulation results, we verify the theoretical analysis and show that the performance of our algorithm.

In this paper, we study a distributed link scheduling problem for device-to-device (D2D) communication with considering the quality-of-service (QoS) requirement and time-varying channel condition of each D2D link. To this end, we first study an optimal centralized link scheduling algorithm maximizing the total average sum-rate. We then abstract the important scheduling principles of the optimal algorithm in order to use them to develop a distributed link scheduling algorithm. In our distributed link scheduling algorithm, we develop a procedure with which D2D links are able to share their degree of QoS satisfaction and channel condition with each other in a distributed manner. By utilizing those information for link scheduling, our link scheduling algorithm is able to satisfy the QoS requirement of each D2D link while achieving throughput improvement by exploiting time-varying channel condition of each D2D link in a distributed manner.

We study a distributed link scheduling problem for device-to-device (D2D) communication considering the quality-of-service (QoS) requirements and time-varying channel conditions of D2D links. To this end, we first study an optimal centralized link scheduling problem maximizing the total average sum-rate while satisfying the QoS requirements of D2D links. We then abstract the important scheduling principles of the optimal link scheduling, i.e., giving more chance to be scheduled to the links which have a good channel condition and do not satisfy the QoS requirement, in order to utilize them to develop distributed link scheduling algorithms. With the scheduling principles, we develop a procedure with which D2D links can share their degree of QoS unsatisfaction and channel condition with each other and generate their scheduling priorities according to the shared information in a distributed manner. We also develop a novel distributed link scheduling criterion with which D2D links determine their link scheduling. By using them, we propose distributed link scheduling algorithms, QCLinQ and QC2LinQ, which have significantly smaller signaling overhead and low computational complexity compared with the centralized optimal link scheduling algorithm. Moreover, they closely meet the QoS requirements of D2D links while achieving significant sum-rate improvement over conventional distributed algorithms.

In this paper, we study joint topology management and energy cooperation in cellular networks. The topology management scheme transfers users (traffic) from one base station (BS) to another BS by adjusting the cell- size of the BSs, and the energy cooperation scheme transfers harvested energy from one BS to another BS. We first formulate a joint topology management and energy cooperation problem which aims at minimizing the total on-grid energy consumption while satisfying the quality-of-service (QoS) requirement of each user. Then, by solving the problem, we develop a joint topology management and energy cooperation scheme. Topology management and energy cooperation are different approaches to save the on-grid energy consumption, and each scheme is effective to save the on-grid energy consumption in different environments such as different traffic and weather conditions. Through the simulation results, we show which scheme is more effective to save the on-grid energy consumption considering various environments. In addition, we also show that our joint scheme is most effective to deal with the environment which changes dynamically.